A high-performance C++ and CUDA inference engine for single-GPU execution, optimized for Qwen checkpoints and RTX 5090 hardware across text, image, and video with a compatible API.